Soksoo Database

    Keywords: Software

Soksoo was a program that read games of go in sgf format, and created a sort of joseki/trick move database viewable with an sgf browser. It was intended for use with amateur games, as they are readily available, and contain the sort of mistakes amateurs need to learn from. Soksoo (or soksu) was derived from Korean language and means 'not joseki'. Sadly the program disappeared[1] from the net.

Table of contents


(complete description, as found on

Soksoo is a program that reads games of go in sgf format, and creates a sort of joseki/trick move database viewable with an sgf browser. It is intended for use with amateur games, as they are readily available, and contain the sort of mistakes amateurs need to learn from.

Average rank, minimum rank, maximum rank and raw count of the players playing a move are given at each node in the resulting sgf file, which can be viewed as a sort of amateur joseki dictionary. This enables one to find a relatively strong refutation to a nonjoseki (soksoo in Korean, shingata in Japanese - anyone know the Chinese?) move.

To build the real thing, you have to get the games from KGS, and last I heard, you had to ask around to get them. However, I've saved you the trouble by including a copy of soksoo.sgf from a big run ("make huge") in the distribution. You can still do "make tiny", but you can't do "make small", "make medium", "make large" or "make huge" out of the box.

Soksoo rhymes with the Japanese "Moku". The name was kindly suggested by sbchoi of KGS.

Soksoo makes pretty heavy demands on an sgf editor.

  • It doesn't appear to work with gGo 0.4.1. gGo gives an error while reading the file.
  • It does appear to work with cgoban 2.5.4, though it's a bit hard to navigate this kind of deep tree with cgoban. Same goes for cgoban 1.9.14. I suggest not dragging the scrollbar - you'll probably just lose your place. The slow arrows at the ends of the scrollbar in the variation tree window are apparently the best option.
  • qGo reportedly works fine. On my system (Redhat 9), I had to change the final link to use c++ instead of gcc, otherwise a bunch of c++-specific methods came up undefined. The resulting binary segfaulted until I monkied with my XF86Config, but now it does appear to work well. Variations are identified with a stone - nice. Update: There's may a problem when there are two moves at the same spot with different colors. Particularly the low approach on a 3-4 is bad. However, inspection with cgoban 1.x preliminarily suggests this may be a problem with soksoo itself. More later.
  • RubyGo appears to work pretty well, except the scrollbar for the list of variations doesn't work - and soksoo has enough variations that the scrollbar is needed. It presents a list of variations to choose from, identified only by a letter (not a position on the board), but it seems to do the best of the sgf editors I've tried so far.
  • ManyFaces reportedly just gives an error when reading the file.
  • Goban ( reportedly works, though my use of AB instead of B may be muddling things for the variation selector a bit.
  • I'm interested in hearing (here mail address) about which other SGF editors it does/doesn't work with.

The code is GPL licensed. The doc it generates is FDL licensed.


  1. Stones that are captured, are not removed from the board in this release
  2. Corners are examined in isolation. You have no information about ladders and other ways other corners impact the corner under study and vice versa. I'm sure soksoo will be heavily criticized for this, but I still find it useful. If you do too then great; if not, that's fine too.
  3. Variations probably should be sorted by average strength, or perhaps max strength. Unfortunately, ISTR hearing that some SGF editors display variations in the opposite order of others.


Discussion and Research

Calvin: The data from KGS is a bit dirty (it seems to include game reviews and problem examples) but I think it would be great if SL could start going through this to identify and analyze variations that are common but are not played by players stronger than say, 10k, or 2dan, or whatever; i.e., common joseki mistakes. Maybe someone with access to recent KGS or IGS data (or games from another server) could process it and create a new file. I'd be willing to take a stab at it, and fix the program and output if necessary. Of course, I'd need strong players to explain the refutations.

Phelan: SDK Game Joseki? is an attempt to do this, I think.

Stromberg?: Soksoo ignores SGF inputs with variations and handicaps.

Soksoo Database last edited by on September 13, 2017 - 06:12
RecentChanges · StartingPoints · About
Edit page ·Search · Related · Page info · Latest diff
[Welcome to Sensei's Library!]
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Login / Prefs
Sensei's Library